Deep Fake

website Link:

https://deepfake-3ux1.onrender.com/ Deep Fake is a Django-based AI content detection studio for checking whether an image looks fake and whether an essay looks AI-generated. Real vs fake image detection using a custom CNN served with Django.

What It Does

• Image detector powered by a custom PyTorch CNN.
• Text detector powered by a PyTorch RNN with TF-IDF features.
• Shared website shell with a navbar and room for more detectors later.
• Clean, responsive UI with live image preview and separate pages for each tool.
• Volatile uploads only. Nothing is persisted after inference.

Website Structure

• Home introduces the product and links to each detector.
• Image Check accepts an image upload and returns a real/fake verdict.
• Text Check accepts an essay and returns an AI/human-written verdict.

Project Architecture

• deepfake_web/
  • Django project settings, URLs, and WSGI entry point.
• predictor/
  • Views, forms, and model-loading helpers.
• templates/
  • base.html shared shell.
  • home.html landing page.
  • image_detector.html image workflow.
  • text_detector.html essay workflow.
• static/predictor/
  • CSS for the entire site.

Model Architecture

Image Model

Input: 3 x 32 x 32

Conv2d(3 -> 32, kernel=3, padding=1)
ReLU
MaxPool2d(2)

Conv2d(32 -> 64, kernel=3, padding=1)
ReLU
MaxPool2d(2)

Conv2d(64 -> 128, kernel=3, padding=1)
ReLU
MaxPool2d(2)

Flatten
Linear(2048 -> 256)
ReLU
Linear(256 -> 1)
Sigmoid

Text Model

Input: TF-IDF features with max_features=5000

TF-IDF vectorizer
RNN(input_size=5000, hidden_size=128)
Linear(128 -> 1)
Sigmoid

Inference Flow

1. User opens the relevant page from the navbar.
2. Django receives the image or text input.
3. The input is normalized to match training.
4. The model runs in evaluation mode with torch.no_grad().
5. The app returns a human-readable verdict and a score.

Local Setup

1) Activate the virtual environment

The project expects the venv at: /Users/aayannoori/Desktop/django/.venv

2) Install dependencies

/Users/aayannoori/Desktop/django/.venv/bin/python -m pip install -r requirements.txt

3) Run the server

/Users/aayannoori/Desktop/django/.venv/bin/python manage.py runserver

Open: http://127.0.0.1:8000/

Deployment Notes

• Gunicorn is the production server.
• WhiteNoise serves static files.
• PostgreSQL is preferred, with SQLite fallback if Postgres is unavailable.

Important Files

• predictor/services.py
• predictor/views.py
• templates/base.html
• templates/image_detector.html
• templates/text_detector.html

Missing Text Artifact

The essay detector also expects:

• best_rnn_weights.pth
• tfidf_vectorizer.pkl

The weights file is already in the repo. If the vectorizer is not present yet, the page still renders and shows a setup hint instead of crashing.